Inspect a Dataset
Inspector Workflow
Section titled “Inspector Workflow”Inspector starts with dataset selection and empty-state troubleshooting, then moves into overview cards, balance analysis, per-class controls, stats, recommendations, and readiness checks.
Labels Ready 4 classes Metadata Ready source OIDs present Balance Check review minority class
Class balance preview
Example modulation classes before and after curation checks.
- Before
- After
Data table
| Series | BPSK | QPSK | 8PSK | 16QAM |
|---|---|---|---|---|
| Before | 32 slices | 18 slices | 12 slices | 9 slices |
| After | 28 slices | 27 slices | 25 slices | 24 slices |
Model comparison placeholder
Compare candidate metrics before choosing an export artifact.
- Validation score
Data table
| Series | Tiny | Base | Tuned |
|---|---|---|---|
| Validation score | 0.71 score | 0.79 score | 0.86 score |
What to Fix Before Training
Section titled “What to Fix Before Training”Inspector troubleshooting
Imbalanced classes
Collect, synthesize, or curate more examples for minority classes before trusting comparisons.
Missing metadata
Check source OIDs, label names, qualifiers, sample shape, and split fields.
Too few examples
Avoid launching a run when each class has only a handful of slices unless the task is explicitly a smoke test.
Invalid split assumptions
Confirm that train and validation examples do not leak from the same source interval when independence matters.
Next steps
Section titled “Next steps”- Train the model — When inspection passes, continue to Train a Model to configure and launch a Model Builder run.
- Re-curate if needed — If issues are found, return to Curation and Labeling to adjust the slicer or qualifier settings.